1: \begin{abstract}

2: The Truncated Nonsmooth Newton Multigrid (TNNMG) method is a well-established method for the solution of strictly convex

3: block-separably nondifferentiable minimization problems.  It achieves multigrid-like performance even for non-smooth

4: nonlinear problems, while at the same time being globally convergent and without employing

5: penalty parameters.  We show that the algorithm can be applied to the primal problem of classical linear elastoplasticity

6: with hardening.  Numerical experiments show that the method is considerably faster

7: than classical predictor--corrector methods.  Indeed, solving an entire increment problem with

8: TNNMG takes less time than a single predictor--corrector iteration for the same problem.

9: Since the algorithm does not rely on differentiability of the objective functional, nonsmooth

10: yield functions like the Tresca yield function can be easily incorporated.

11: The method is closely related to a predictor--corrector scheme with a consistent tangent predictor and

12: line search.  We explain the

13: algorithm, prove global convergence, and show its efficiency using a standard benchmark from the literature.

14: \end{abstract}

15: